1. Technical Field
The present invention relates to a monitoring device, a reliability calculation program, and a reliability calculation method, which process images captured with a plurality of imaging devices having different imaging areas, detect the image in which a detection target person is captured, and output a detection notification about the detection target person of the detected image.
2. Related Art
Nowadays, monitoring cameras are placed at various sites, such as an airport, a station, a shopping center, and a street corner. An image, in which a detection target person who is a person on a wanted list is captured, is detected by performing face recognition in which a face of a person in the image captured with the monitoring camera is matched against a face of the detection target person (for example, see Japanese Unexamined Patent Publication No. 2004-62560).
In this kind of monitoring device, the monitoring device notifies a related organization of a detection notification including a site where the image is captured, date and time, and the imaged detection target person when the image in which the detection target person is captured is detected.
In the related organization, an agent searches the detection target person based on the detection notification.
However, in the face recognition, a degree of similarity between a feature quantity of the face of the person in whom the image is captured with the monitoring camera and a feature quantity of the face of the detection target person is calculated, and the person in whom the image is captured with the monitoring camera is determined to be the detection target person when the calculated degree of similarity exceeds a predetermined criterion.
In the face recognition, it is necessary that the criterion for the degree of similarity be extremely increased in order to reduce a probability of mistakenly recognizing another person in whom the image is captured with the monitoring camera as the detection target person (false match rate) to zero. At the same time, the probability of mistakenly recognizing the detection target person in whom the image is captured with the monitoring camera as another person (false rejection rate) increases with increasing criterion for the degree of similarity. Accordingly, in the face recognition, it is necessary that the probability of mistakenly recognizing the detection target person in whom the image is captured with the monitoring camera as another person (false rejection rate) be brought close to one in order to reduce a probability of mistakenly recognizing another person in whom the image is captured with the monitoring camera as the detection target person (false match rate) to zero, and it is difficult to balance the probabilities in terms of usefulness.
Therefore, the number of detection notifications due to the false detection of the detection target person (a notification provided by mistakenly recognizing another person as the detection target person) increases with increasing number of monitoring cameras used to detect the detection target person. For example, even if the probability of mistakenly recognizing another person in whom the image is captured with the monitoring camera as the detection target person is 0.1%, the detection notification is provided once per 10 minutes due to the false recognition of another person as the detection target person in the monitoring camera that captures images of 100 walking persons per minute. In the case that 100000 monitoring cameras are used (similarly, in the case that one monitoring camera captures images of 100 walking persons per minute), the detection notification is provided 10000 times per minute due to the false recognition of another person as the detection target person. That is, in the related organization, although the agent should search the detection target person with respect to the detection notifications, which are provided 10000 times per minute due to the false recognition of another person as the detection target person (substantially useless searches), it is difficult to secure human resources, and unfortunately it is ideal. Due to a shortage of the human resources, sometimes the detection target person is not searched with respect to the detection notification that is provided by correctly recognizing the detection target person as the detection target person.
The number of detection notifications due to the false recognition of another person as the detection target person can be decreased by decreasing the number of monitoring cameras used to detect the detection target person. However, a potential to capture the image of the detection target person (namely, a potential to be able to detect the detection target person) is decreased.